Unsupervised clustering of visitors based on click logs | July - March 2014.
As a part of my Masters thesis I am currently trying to cluster user sessions using the click paths of a user on a web site in an unsupervised manner. So far, I have come up with a theoretical model to cluster the sessions. Further work would involve coming up with useful insights about the clusters formed on various real world data sets. My thesis was awarded K.L. Chopra award for the best demonstrable project in the session 2013 - 2014.
Guide: Professor Somesh Kumar,Department of Mathematics, IIT Kharagpur
Recommender System : Deep Learning | May 2012 - July 2012
This was the summer internship I pursued under Professor Yoshua Bengio of University of Montreal. It was my first exposure to Deep learning. After exploring few initial concepts, I built a Recommender system which would suggest users a list of gaming inventory to purchase based on his gaming profile. This was achieved by a Supervised Multi Layer Perceptron model using deep learning techniques. The model out-performed the existing model that was used for recommendations earlier significantly in terms of precision.
Guide: Professor Yoshua Bengio. Head of Machine Learning Laboratory, University of Montreal, Canada.
Webpage classification : Machine Learning| May 2011 - July 2011
This project was pursued at Chitika Labs, an online ad delivery platform. This was my first exposure to applied machine learning. In this project I built a multi-class classifier for a set of web paged .This was achieved using Co-training model combined with a Naive Bayes Classifier. The two views that were considered for the Co-training were the text on a web page and the hyperlinks on it. The model gave satisfactory results on the test sample.
Internship Mentor: Warren Noronha, CTO, Chitika India.